biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Bugfix][Model] Fix DeepSeek-OCR TensorSchema crash on empty images_crop (#36670)

Browse Source 
Signed-off-by: István Ketykó <istvan.ketyko@gmail.com>
This commit is contained in:
István Ketykó
2026-03-12 08:35:54 +01:00
committed by GitHub
parent 9fe404ed04
commit 00726c74c9
2 changed files with 135 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/model_executor/models/deepseek_ocr.py
View File

@@ -452,10 +452,7 @@ class DeepseekOCRForCausalLM(nn.Module, SupportsMultiModal, SupportsPP, Supports
# support arbitrary resolutions via pos-encoding interpolation,
# so Tiny/Small/Base/Large variants all work with the same weights.
base_size = pixel_values.shape[-1]
if images_crop is not None and images_crop.numel() > 0:
image_size = images_crop.shape[-1]
else:
image_size = base_size
image_size = images_crop.shape[-1] if images_crop is not None else base_size
return DeepseekOCRImagePixelInputs(
type="pixel_values",